deepseek/deepseek-reasoner

Aliases: r1, deepseek-r1

$0.55

Input / 1M

$2.19

Output / 1M

64K

Context

30 t/s

Speed

Public benchmark scores

Sourced from each provider's published numbers. Verify before quoting.

Quality index

MMLU

90.8

GPQA

71.5

HumanEval

86.4

MATH

97.3

SWE-bench

—

Arena Elo

—

Sources: deepseek-r1-paper

Capabilities

streamingthinking

Use deepseek-reasoner via Relay

Configure the model alias in YAML, then call it from Python.

YAML

# models.yaml
version: 1
models:
  r1:
    target: deepseek/deepseek-reasoner
    credential: $env.DEEPSEEK_API_KEY

Python

from relay import Hub

async with Hub.from_yaml("models.yaml") as hub:
    resp = await hub.chat(
        "r1",
        messages=[{"role": "user", "content": "Hello"}],
    )
    print(resp.text, resp.cost_usd)

pip install ai5labs-relay · full docs on GitHub

Compare with

vs openai/o3-mini vs anthropic/claude-sonnet-4-5 vs anthropic/claude-opus-4-5 vs deepseek/deepseek-chat